Non-calcified coronary atherosclerotic plaque characterization by dual energy computed tomography

Area of Science:

• Cardiovascular Imaging
• Medical Artificial Intelligence
• Radiology

Background:

• Coronary heart disease (CHD) is a leading global cause of mortality.
• Atherosclerosis, characterized by coronary artery plaque buildup, underlies CHD.
• Plaque vulnerability, linked to lipid core and fibrous cap composition, predicts acute coronary syndrome.

Purpose of the Study:

• To leverage dual-energy computed tomography (DECT) for non-invasive atherosclerotic plaque characterization.
• To enhance the precision of classifying fibrous versus lipid plaques using DECT-derived data.

Main Methods:

• Supervised machine learning models were trained on pixel values from DECT monochromatic X-ray and material basis pair images.
• The study considered interactions between pixel values from different image types for improved classification.
• Organic phantom plaques within a beating heart phantom served as ground truth for model training.

Main Results:

• Support vector machines, artificial neural networks, and random forests demonstrated accurate plaque classification.
• The proposed DECT-based approach showed efficacy in both phantom and patient datasets.
• Machine learning models effectively utilized multi-modal DECT data for plaque differentiation.

Conclusions:

• DECT provides valuable attenuation data for non-invasive atherosclerotic plaque characterization.
• Supervised learning algorithms can accurately classify plaque components using DECT imaging.
• This approach holds promise for early identification and risk stratification of coronary heart disease.